OpenAI Unveils Jalapeño, Its First Custom AI Chip Built With Broadcom

The AI industry is burning through compute faster than chipmakers can comfortably supply it. Memory shortages, rising component costs, and skyrocketing demand for AI infrastructure have already forced companies to rethink how they build the next generation of data centers. OpenAI now appears ready to tackle that problem itself.

Today, the company officially unveiled Jalapeño, its first custom AI chip developed in partnership with Broadcom. Rather than relying entirely on third-party hardware, the ChatGPT maker is now stepping further into the silicon business with a chip specifically built for large language model inference.

OpenAI wants more control over AI hardware, and it’s first product is here

While most people know OpenAI through ChatGPT, the company has increasingly been investing behind the scenes in the infrastructure powering those services. Jalapeño is a major part of that strategy. According to OpenAI, the chip was designed from the ground up around real-world AI workloads rather than being adapted from older chip designs.

The company says engineering samples are already running production-level machine learning tasks, including GPT-5.3-Codex-Spark workloads, while operating at target power and frequency levels.

Early testing apparently also shows performance-per-watt improvements over current state-of-the-art AI chips. OpenAI hasn’t shared final benchmark numbers yet. We expect a detailed later down the line.

As for Jalapeño, Broadcom handled silicon implementation and networking technologies, while Celestica contributed board, rack, and system-level integration.

OpenAI says Jalapeño moved from initial design to manufacturing tape-out in just nine months, a pace the company believes is among the fastest ever achieved for an advanced AI chip. OpenAI has further confirmed that Jalapeño is only the first entry in a multi-generation platform expected to power gigawatt-scale deployments with partners starting in 2026.

For a company once known solely for AI models, OpenAI is increasingly building everything underneath them too. Models, products, infrastructure, and now custom silicon are all becoming part of the same playbook.